A fingerprint library indoor positioning method assisted by intelligent reflective surface
Through the indoor positioning method of the fingerprint library assisted by intelligent reflective surface, neural network training and channel state information are used to solve the problem of positioning performance degradation caused by complex indoor environments, and efficient positioning is achieved when the visual line link is blocked.
Patent Information
- Application Number
- CN202211057712.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-31
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2042-08-31
AI Technical Summary
In the case of many obstacles in the indoor environment, the line-of-sight link between the base station and the user is blocked, resulting in increased channel path loss, reduced positioning performance, and significantly increased errors.
Using the indoor positioning method of fingerprint library assisted by intelligent reflective surface, by establishing a cascaded channel model between the user, intelligent reflective surface and base station, using multi-layer feedforward neural network training and error reverse propagation algorithm, combining the range of sight links and levels The channel status information of the connecting channel and the general channel is positioned.
Maintain good positioning performance in complex indoor environments, reduce positioning errors, improve positioning efficiency, and achieve accurate positioning while the line-of-sight link is blocked.
Smart Images

Figure CN115474270B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of wireless communications, and in particular relates to an indoor positioning method of a fingerprint library assisted by an intelligent reflective surface. Background Art
[0002] Smart reflectors are a new type of artificial electromagnetic material, which consists of a large number of reflective units with the same structure. From a microscopic perspective, the reflection coefficient (phase, amplitude) of each reflective unit can be artificially controlled to make it independently exert a controllable influence on the incident electromagnetic wave; from a macroscopic perspective, all reflective units can be coordinated to change the number, direction, scattering degree, etc. of the reflected beam.
[0003] In a multiple input single output (MISO) communication environment, the traditional least squares (LS) method can be used to estimate the total channel and the channel state information (CSI) of the line-of-sight link and the base station-intelligent reflecting surface-user cascade link can be obtained by using channel decomposition and sub-channel estimation methods.
[0004] In general, indoor positioning is divided into distance measurement-based and distance-independent positioning methods, of which a typical method of the latter is the fingerprint library algorithm. Common methods are based on the received signal strength indicator (RSSI), but when there are many obstacles in the indoor environment, the line-of-sight link between the base station (BS) and the user may be blocked, the channel path loss increases, the positioning performance is affected, and the error increases significantly. Summary of the invention
[0005] The purpose of the present invention is to provide a fingerprint library indoor positioning method assisted by an intelligent reflective surface to solve the technical problem that when there are many obstacles in the indoor environment, the line-of-sight link between the base station and the user may be blocked, the channel path loss may increase, the positioning performance may be affected, and the error may increase significantly.
[0006] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows:
[0007] A fingerprint library indoor positioning method assisted by an intelligent reflective surface comprises the following steps:
[0008] Step 1: Create a fingerprint database;
[0009] Define the line-of-sight link channel between the user and the base station as h d , the channel between the user and the smart reflective surface is h r, the channel between the intelligent reflection surface and the base station is g, the above channel models are all single-path time-invariant channel models, then G = g*diag(h r ) is defined as the cascade channel between the user, the smart reflective surface, and the base station;
[0010] The base station is equipped with multiple antennas, the user is equipped with a single antenna, the intelligent reflection surface is equipped with multiple reflection units, and multiple fixed coordinate points are pre-selected indoors as reference points of the fingerprint library; the intelligent reflection surface is turned on, the user sends pilots in multiple time slots at each fixed coordinate point, and the base station obtains the channel state information CSI of the total link, which is recorded as CSI-1; the intelligent reflection surface is turned off, the user sends pilots in multiple time slots at each fixed coordinate point, and the channel state information CSI of the link that does not pass through the intelligent reflection surface is obtained, which is recorded as CSI-2; the indirect channel state information CSI value passing through the intelligent reflection surface is obtained by subtracting CSI-1 from CSI-2, which is recorded as CSI-3; the steps of obtaining CSI-1, CSI-2, and CSI-3 are repeated in each time slot, so that the result satisfies the statistical probability distribution; the amplitude information of the above three types of channel state information CSI is obtained, and the link P that does not pass through the intelligent reflection surface is established respectively. d , link P reflected by the smart reflective surface r and the total link P to obtain the channel state information CSI fingerprint library;
[0011] Step 2: Obtain neural network training samples;
[0012] Q coordinate positions are randomly selected indoors, and a user sends a pilot signal at each position. The channel state information CSI of the three links mentioned in step 1 is obtained at the base station by switching the intelligent reflective surface, and the three types of fingerprint information in the channel state information CSI fingerprint library obtained in step 1 are subtracted respectively. The obtained differences are squared, and then weighted summed to obtain the coordinate positions of multiple reference points with the largest sum, which are used as the input of the neural network, and the actual coordinate parameters of the point are used as the output; step 2 is repeated Q times to obtain Q groups of neural network training samples;
[0013] Step 3: Train the neural network;
[0014] Step 4: In the online stage, the user sends a positioning request to the base station;
[0015] The base station responds and prepares to receive the pilot sent by the user, and obtains the channel state information CSI of the three links using the method in step 2 and takes the modulus value;
[0016] Both step 1 and step 2 obtain the channel state information of the three links. The difference is that in step 1, some fixed points are selected in advance, at which the user sends pilot signals, and the base station receives and obtains the CSI information. There is only one more step to store them in the fingerprint library. Step 2 repeatedly obtains the CSI information of the three links at random points, and the acquisition method is similar to step 1. The method used in step 2 here is that the user sends pilot information to the base station at the current location, and the base station obtains the channel state information of each link;
[0017] Step 5: Match the channel modulus values obtained in step 4 with the fingerprint library to obtain the coordinate positions of the most matching points;
[0018] The possible coordinate position points are input into the neural network obtained in step 3 for calculation to obtain the estimated position information, and the positioning is completed.
[0019] Furthermore, step 3 specifically includes the following steps:
[0020] Perform multiple rounds of neural network training on the xyz three-dimensional space coordinate values, and estimate the actual position information based on the given possible position information; compare the estimated position with the actual position of each point in step 2. If the average positioning error of the test point is less than the threshold D, save the set of neural networks, otherwise repeat this step.
[0021] Furthermore, the weight ratio between the channel state information CSI of the link passing through the smart reflective surface, the channel state information CSI of the link not passing through the smart reflective surface and the channel state information CSI of the total link is not unique. The positioning performance is good when the weight of the link not passing through the smart reflective surface is greater than the sum of the weights of the other two.
[0022] Furthermore, by reducing the number of fingerprint database points in step 1 and the number of training sample groups Q in step 2, good positioning performance can be maintained while improving positioning efficiency.
[0023] Furthermore, in step 3, the neural network uses a BP neural network or a convolutional neural network CNN. The purpose of neural network training is to find the relationship between the positioning coordinates and the possible position coordinates, and is not limited to using a certain neural network or taking other methods to find the mapping relationship between the two.
[0024] The fingerprint library indoor positioning method assisted by an intelligent reflective surface of the present invention has the following advantages:
[0025] Traditional indoor positioning algorithms will lead to reduced positioning performance and increased positioning errors when the indoor environment is complex. This method uses intelligent reflective surfaces to assist in establishing a fingerprint library, comprehensively utilizes the channel state information of the line-of-sight link channel, cascade channel and total channel as a reference, and uses a multi-layer feedforward neural network trained by the error back propagation algorithm to assist in training for positioning. It can maintain good positioning performance in complex indoor environments and when the line-of-sight link is blocked. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 This is a schematic diagram of scene distribution of the present invention;
[0027] Figure 2 It is a cumulative distribution function curve diagram of positioning errors of various methods of the present invention;
[0028] Figure 3 The present invention is an indoor positioning error color difference gradient map;
[0029] Figure 4 The relationship diagram between the positioning power consumption reduction (number of reference points in the fingerprint library) and the positioning performance of the present invention;
[0030] Figure 5 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION
[0031] In order to better understand the purpose, structure and function of the present invention, the following is a further detailed description of the fingerprint library indoor positioning method assisted by an intelligent reflective surface of the present invention in conjunction with the accompanying drawings.
[0032] A fingerprint library indoor positioning method based on neural network intelligent reflective surface assistance includes the following steps:
[0033] Step 1: Establish a fingerprint database. Define the line-of-sight link channel between the user and the base station as h d , the channel between the user and the smart reflective surface is h r , the channel between the intelligent reflector and the base station is g, the above channel models are all single-path time-invariant channel models, then G = g*diag(h r) is defined as the cascade channel between the user, the smart reflective surface, and the base station. The base station is equipped with multiple antennas, the user is equipped with a single antenna, and the smart reflective surface is equipped with multiple reflective units. Pre-select multiple fixed coordinate points in the room as fingerprint library reference points, and turn on the smart reflective surface. The user sends a pilot signal in the room, and the base station receives the channel state information (Channel State Information, CSI) of the total link at the fixed coordinate point, recorded as CSI-1. Turn off the smart reflective surface, repeat the above operation, and obtain the CSI-2 of the link channel that does not pass through the smart reflective surface. The CSI-3 value of the channel passing through the smart reflective surface can be calculated from the above two. Repeat this step in each time slot so that the result satisfies the statistical probability distribution. Take the amplitude information of the CSI, and establish the link P that does not pass through the smart reflective surface respectively. d , link P reflected by the smart reflective surface r and the fingerprint library of CSI information of the total link P.
[0034] Step 2: Get neural network samples. Randomly select Q coordinate positions indoors. At each position, the user sends a pilot signal, and the CSI information of the three links mentioned in step 1 is obtained at the base station through the switch intelligent reflector and data processing. The CSI information is weighted with the channel amplitude information modulus value of the CSI of the reference point in the fingerprint library obtained in step 1, and a comprehensive comparison is performed to obtain the coordinate positions of the most matching points, which are used as the input of the neural network, and the actual coordinate parameters of the point are used as the output. Repeat the above process Q times to obtain a neural network sample with possible coordinate positions as input and actual coordinate positions as output.
[0035] Step 3: Train the neural network. Perform multiple rounds of neural network training on the xyz three-dimensional space coordinate values, and estimate the actual position information from the given possible position information. Compare the estimated position with the actual position of each point in step 2. If the average positioning error of the test point is less than the ideal value D, save the set of neural networks, otherwise repeat this step.
[0036] Step 4: In the online phase, the user sends a positioning request to the base station. The base station responds and prepares to receive the pilot sent by the user, and obtains the CSI information of the three links using a method similar to that in step 2.
[0037] Step 5: Match the channel modulus values obtained in step 4 with the fingerprint library to obtain the coordinate positions of the most matching points. Input the possible coordinate position points into the neural network obtained in step 3 for training to obtain the estimated position information, and the positioning is completed.
[0038] A method for indoor positioning by using a fingerprint library assisted by an intelligent reflective surface, which comprehensively utilizes the channel state information of the line-of-sight link channel, the cascade channel and the total channel as a reference, and simultaneously uses a multi-layer feedforward neural network trained by an error back propagation algorithm to assist in training for positioning, and can maintain good positioning performance even in complex indoor environments and when the line-of-sight link is blocked. It specifically includes the following steps:
[0039] Embodiment 1:
[0040] The base station has N t The intelligent reflective surface has M reflective units, and a single user has an omnidirectional antenna. The room size is 11m×11m, and points are taken every 0.5m as fingerprint library reference points, with a total of N=441 reference points. A coordinate system is established with the center of the room as the origin, the base station position (-1m, -5.5m, 1.5m), and the intelligent reflective surface (-1m, -5.5m, 3m). The position distribution of users, base stations, and intelligent reflective surfaces in the room is as follows Figure 1 In this example, N is set t =8, M = 64. The line-of-sight link channel between the user and the base station is modeled as:
[0041]
[0042] in is the arrival angle of the line-of-sight link signal at the base station, L d is the number of channel paths, is the steering vector, α i (i=0,1,…,L d -1) indicates the corresponding path gain.
[0043] The channel between the smart reflective surface and the base station is modeled as a Rice channel:
[0044]
[0045] K 1 represents the Rice factor of g, represents the line-of-sight link component and can be expressed as is the steering vector, θ AOA,1 represents the angle of arrival at the base station, θ DOA,1 Represents the departure angle at the smart reflective surface. represents the non-line-of-sight link component, each element of which obeys an independent complex Gaussian distribution with a mean of 0 and a variance of 1. is α N The conjugate transpose of .
[0046] The channel between the user and the smart reflective surface is modeled as a Rice channel:
[0047]
[0048] K 2 Represents h r The Rice factor, represents the line-of-sight link component and can be expressed as is the steering vector, θ AOA,2 represents the arrival angle at the smart reflector, θ DOA,2 Represents the departure angle at the user. represents the non-line-of-sight link component, each element of which obeys an independent complex Gaussian distribution with a mean of 0 and a variance of 1.
[0049] In this example, the reflection coefficient vector v of the smart reflective surface 0 The amplitude is set to 1, and the phase is set to a random angle in [0,π). The channel Rice factor is set to K 1 =K 2 =10, the arrival angle at the base station, the arrival angle and the departure angle of the smart reflector are all set to random angles that obey a uniform distribution in [0,π).
[0050] When the intelligent reflector is not turned on, the user sends the p-round pilot x i (i=1,2,...p), receiving line-of-sight link channel h in the fingerprint database d =[h d1 ,h d2 ,...,h dp ], where h di (i=1,2,...p)=[h di1 ,h di2 ,...,h diN ] indicates that when sending the i-th round of pilot signals, the base station receives the signal vector of N reference coordinate points, where The signal vectors of each antenna line-of-sight link of the jth fingerprint library reference point are received during the i-th round of pilot. In this example, p=20.
[0051] Turn on the smart reflector, and the user sends a p-wheel pilot x i (i=1,2,...p), the base station receives the total link channel G=[G 1 ,G 2 ,...,G P ], where G i (i=1,2,...p)=[G i1 ,G i2 ,...,G iN ] indicates that when sending the i-th round of pilot signals, the base station receives N reference coordinate point signal vectors, where The signal vector of each antenna line-of-sight link of the jth fingerprint library reference point is received during the i-th round of pilot.
[0052] Calculate the CSI module information respectively to obtain the module information P of the line-of-sight link d =[P d1 ,P d2 ,...,P dN ],in The total channel modulus information P = [P 1 ,P 2 ,...,P N ],in Therefore, the modulus information P of the cascade channel through the smart reflector can be calculated. r =[P r1 ,P r2 ,...,P rN ], where P rj (j=1,2,...N)=P j -P dj At this point, the storage of the three channel information in the fingerprint library is complete.
[0053] Step 2: Configure the smart reflective surface phase shift coefficient matrix as μ and define At the same time, define the matrix The base station adjusts the phase shift to μ in each of L (L≥M+1=65) time slots. l And send a round of pilot signals, select Q coordinate positions [NetLoc 1 ,NetLoc 2 ,...,NetLoc Q ]
[0054] NetLoc i (i=1,2,...,Q)=[NetLoc 1x ,NetLoc 1y ,NetLoc 1z ] is the coordinate vector of the i-th point, and the signal is received at these Q points. In this example, Q = 1000 defines the total channel:
[0055] H e,l =h d +G
[0056] Using traditional channel estimation such as LS method, we can get L estimates of the total channel. Through channel decomposition and subchannel estimation methods,
[0057] When ∑ satisfies certain conditions [1], we get the line-of-sight link channel h dEstimation of the concatenated channel G:
[0058]
[0059] Repeat this step u times to satisfy the statistical probability distribution, h dnet =[h dnet1 ,h dnet2 ,...,h dnetu ],h dneti (i=1,2,...u)=[h dneti1 ,h dneti2 ,...,h dnetiQ ] represents the signal vector received by Q sample reference coordinate points when sending the i-th round pilot, where is the line-of-sight link signal vector received by each antenna. Then the channel modulus information of the line-of-sight link can be expressed as = [P dnet1 , P dnet2 ,…,P dnetQ ],in Similarly, the modulus information of the cascade channel can be expressed as P rnet =[P rnet1 , P rnet2 ,…,P rnetQ ],in The total channel modulus information = [P net1 , P net2 , ..., P netN ],in
[0060] Based on the various CSI information obtained previously, the matching matrix Mat of the jth (j=1, 2, ..., Q) point is defined as j =[Mat j1 , Mat j2 , …, Mat jN ], where Mat ji (i=1, 2, ...N)=λ 1 (P dnetj -P di ) 2 +λ 2 (P rnetj -P ri ) 2 +λ 3 (P netj -P i ) 2 , for Mat j Sort them, take m = 5 values from small to large, and get the coordinates of the reference point of the fingerprint library corresponding to these m values [Loc j1 , Loc j2 ,...Loc jm], Loc ji (i=1, 2, ..., 5)=[Loc jix , Loc jiy , Loc jiz ] is a vector containing the three-dimensional coordinate values of the fingerprint library, and it is used as the input of the neural network, and the coordinates of the jth point are used as the output of the network. Repeat Q times, and the neural network sample acquisition is completed. In this example, λ is selected 1 =4,λ 2 =1,λ 3 =2.
[0061] Step 3: Train the neural network. In this example, the method adopted is to train the neural network for each dimension of the coordinate point. The training sample is [P Loc 1 , P Loc 2 , ..., PLoc m , ALoc], where PLoc i (i=1,2,…m)=[Loc 1i , Loc 2i ,…,Loc Qi ] T , ALoc=[NetLoc 1 , NetLoc 2 , ..., NetLoc Q ] T , the x, y, and z coordinate values in the training samples are trained with the multi-layer feedforward neural network using the error back propagation algorithm. As the test set, the rest is used as the training set, and ten-fold cross validation is performed. The result of each training is the estimated position At this time The actual position of the point is If the average error
[0062]
[0063] If it is less than the threshold D = 0.5m, the network is saved as Net i (i=x, y, z), corresponding to the neural network of each dimensional coordinate, otherwise repeat the training.
[0064] Step 4: In the online phase, the user sends a positioning request to the base station at the current location. The base station responds and prepares to receive the pilot sent by the user, obtains the CSI information of the three links using a method similar to step 2, and calculates the amplitude information and saves it as EstP d , EstP r , Est P. .
[0065] Step 5: Match the channel modulus values obtained in step 4 with the fingerprint library in a similar way to step 2, and obtain the matching matrix Mat = [Mat 1 , Mat 2 , ..., Mat N ], where Mat i (i=1,2,…,N)=λ 1 (EstP d -P di ) 2 +λ 2 (EstP r -P ri ) 2 +λ 3 (EstP-P i ) 2 , sort Mat, take m = 5 values from small to large, and get the coordinates of the fingerprint library reference points corresponding to these 5 values [Loc 1 , Loc 2 ,…,Loc 5 ], Loc i (=[Loc ix , Loc iy , Loc iz ] is a vector containing the three-dimensional coordinate values of the fingerprint library, and it is used as the input of the neural network. The possible coordinate position points are input into the neural network Net obtained in step 3. i (i = x, y, z) to train and obtain the estimated location information Loc 0 =(x 0 ,y 0 , z 0 ), positioning is completed. Figure 2 The cumulative distribution function (CDF) curve of the positioning error is given. In addition to the present invention, two traditional methods are compared, one of which is adaptive weighted ( Figure 2 The method marked as "adaptive weighting" in The positioning result is Another method is to directly select only the coordinate point with the smallest positioning error as the positioning result, ( Figure 2 Marked as "optimal matching point" in the above 1 Compare Loc 0 , Loc′ and Loc 1 And the user's current position, make a cumulative distribution function curve of positioning error as follows Figure 2 It can be seen from the figure that the positioning performance of the present invention is significantly better than the two commonly used methods.
[0066] Figure 3 This is the color difference gradient map of the indoor positioning error at a height of 1.5m.
[0067] Embodiment 2:
[0068] The base station has N t The intelligent reflective surface has M reflective units, and a single user has an omnidirectional antenna. The room size is 11m×11m, and points are taken every 1m as fingerprint library reference points, with a total of N=121 reference points. The coordinate system is established with the center of the room as the origin, the base station position (-1m, -5.5m, 1.5m), and the intelligent reflective surface (-1m, -5.5m, 3m). The indoor positioning scenario is still the same Figure 1 In this example, N is set t =8, M = 64. The line-of-sight link channel between the user and the base station is modeled as:
[0069]
[0070] in is the arrival angle of the line-of-sight link signal at the base station, L d is the number of channel paths, is the steering vector, α i , i=0,1,...,L d -1 indicates the corresponding path gain.
[0071] The channel between the smart reflective surface and the base station is modeled as a Rice channel:
[0072]
[0073] K 1 Represents H 1 The Rice factor, represents the line-of-sight link component and can be expressed as is the steering vector, θ AOA,1 represents the angle of arrival at the base station, θ DOA,1 represents the departure angle at the smart reflective surface. represents the non-line-of-sight link component, each element of which obeys an independent complex Gaussian distribution with a mean of 0 and a variance of 1.
[0074] The channel between the user and the smart reflective surface is modeled as a Rice channel:
[0075]
[0076] K 2 Represents h r The Rice factor, represents the line-of-sight link component and can be expressed as is the steering vector, θ AOA,2represents the arrival angle at the smart reflector, θ DOA,2 Represents the departure angle at the user. represents the non-line-of-sight link component, each element of which obeys an independent complex Gaussian distribution with a mean of 0 and a variance of 1.
[0077] In this example, the reflection coefficient vector v of the smart reflective surface 0 The amplitude is set to 1, and the phase is set to a random angle in [0, π). The channel Ricean factor is set to K 1 =K 2 =10, the arrival angle at the base station, the arrival angle and the departure angle of the smart reflector are all set to random angles that obey a uniform distribution within [0, π).
[0078] When the intelligent reflector is not turned on, the user sends the p-round pilot x i (i=1, 2, ...p), receiving line-of-sight link channel h in the fingerprint database d =[h d1 ,h d2 , ..., h dp ], where h di (i=1, 2, ..., p)=[h di1 ,h di2 , ..., h diN ] indicates that when sending the i-th round of pilot signals, the base station receives the signal vector of N reference coordinate points, where The signal vectors of each antenna line-of-sight link of the jth fingerprint library reference point are received during the i-th round of pilot. In this example, p=20.
[0079] Turn on the smart reflector, and the user sends a p-wheel pilot x i (i=1, 2, ...p), the base station receives the total link channel G=[G 1 , G 2 , ..., G P ], where G i (i=1,2,...p)=[G i1 , G i2 , ..., G iN ] indicates that when sending the i-th round of pilot signals, the base station receives N reference coordinate point signal vectors, where The signal vector of each antenna line-of-sight link of the jth fingerprint library reference point is received during the i-th round of pilot.
[0080] Calculate the CSI module information respectively to obtain the module information P of the line-of-sight link d =[P d1 , P d2 , ..., P dN ],in The total channel modulus information P = [P 1 , P 2 , ..., P N ],in Therefore, the modulus information P of the cascade channel through the smart reflector can be calculated. r =[P r1 , P r2 ,…,P rN ], where P rj (j=1, 2, ...N)=P j -P dj At this point, the storage of the three channel information in the fingerprint library is complete.
[0081] Step 2: Configure the smart reflective surface phase shift coefficient matrix as μ and define At the same time, define the matrix The base station adjusts the phase shift to μ in each of L (L≥M+1=65) time slots. l And send a round of pilot signals, select Q coordinate positions [NetLoc 1 , NetLoc 2 , ..., NetLoc Q ]
[0082] NetLoc i (i=1,2,…,Q)=[NetLoc 1x , NetLoc 1y , NetLoc 1z ] is the coordinate vector of the i-th point, and the signal is received at these Q points. In this example, Q = 1000. Define the total channel:
[0083] H e,l =h d +G
[0084] Using traditional channel estimation such as LS method, we can get L estimates of the total channel. Through channel decomposition and subchannel estimation methods,
[0085] When ∑ satisfies certain conditions [1], we get the line-of-sight link channel h d Estimation of the concatenated channel G:
[0086]
[0087] Repeat this step u times to satisfy the statistical probability distribution, h dnet [h dnet1 ,h dnet2 ,…,h dnetu ],
[0088] hdneti (i=1,2,...u)=[h dneti1 ,h dneti2 , ..., h dnetiQ ] represents the signal vector received by Q sample reference coordinate points when sending the i-th round pilot, where is the line-of-sight link signal vector received by each antenna. Then the channel modulus information of the line-of-sight link can be expressed as = [P dnet1 , P dnet2 , ..., P dnetQ ],in Similarly, the modulus information of the cascade channel can be expressed as P rnet =[P rnet1 , P rnet2 , ..., P rnetQ ],in The total channel modulus information = [P net1 , P net2 , ..., P netN ],in
[0089] Based on the various CSI information obtained previously, the matching matrix Mat of the jth (j=1, 2, ..., Q) point is defined as j =[Mat j1 , Mat j2 , ..., Mat jN ], where Mat ji (i=1, 2, ...N)=λ 1 (P dnetj -P di ) 2 +λ 2 (P rnetj -P ri ) 2 +λ 3 (P netj -P i ) 2 , for Mat j Sort by taking m = 5 values from small to large, and obtain the coordinates of the reference point of the fingerprint library corresponding to these m values [Loc j1 , Loc j2 ,...Loc jm ], Loc ji (i=1, 2, ..., 5)=[Loc jix , Loc jiy , Loc jiz ] is a vector containing the three-dimensional coordinate values of the fingerprint library, and it is used as the input of the neural network, and the coordinates of the jth point are used as the output of the network. Repeat Q times, and the neural network sample acquisition is completed. In this example, λ is selected1 =4,λ 2 =1,λ 3 =2.
[0090] Step 3: Train the neural network. In this example, the method adopted is to train the neural network for each dimension of the coordinate point. The training sample is [PLoc 1 ,PLoc 2 , ..., PLoc m , ALoc], where PLoc i (i=1, 2, ..., m)=[Loc 1i , Loc 2i , ..., Loc Qi ] T , ALoc=[NetLoc 1 , NetLoc 2 , ..., NetLoc Q ] T The x, y, and z coordinate values in the training samples are trained with a multi-layer feedforward neural network using the error back propagation algorithm. points as the test set and the rest as the training set, and a ten-fold cross validation is performed. The result of each training is the estimated position At this time The actual position corresponding to the point is If the average error
[0091]
[0092] If it is less than the ideal value D = 0.5m, then save the network as Net i (i=x, y, z), corresponding to the neural network of each dimensional coordinate, otherwise repeat the training.
[0093] Step 4: In the online phase, the user sends a positioning request to the base station at the current location. The base station responds and prepares to receive the pilot sent by the user, obtains the CSI information of the three links using a method similar to step 2, and calculates the amplitude information and saves it as EstP d , EstP r , Est P. .
[0094] Step 5: Match the channel modulus values obtained in step 4 with the fingerprint library in a similar way to step 2, and obtain the matching matrix Mat = [Mat 1 , Mat 2 , ..., Mat N ], where Mat i (i=1, 2, ..., N)=λ 1 (EstP d-P di ) 2 +λ 2 (EstP r -P ri ) 2 +λ 3 (EstP-P i ) 2 , sort Mat, take m = 5 values from small to large, and get the coordinates of the fingerprint library reference points corresponding to these 5 values [Loc 1 , Loc 2 ,...Loc 5 ], Loc i (i=1, 2, ..., 5)=[Loc ix , Loc iy , Loc iz ] is a vector containing the three-dimensional coordinate values of the fingerprint library, and it is used as the input of the neural network. The possible coordinate position points are input into the neural network Net obtained in step 3. i (i = x, y, z) to train and obtain the estimated location information Loc 0 =(x 0 ,y 0 , z 0 ), the positioning is completed. Compared with the actual position of the user, the cumulative distribution function curve ( Figure 4 Marked as “121 fingerprint library reference points” in the figure). Figure 4 The other two curves are the cumulative distribution function curves of using the method of the present invention and the adaptive weighting method in Example 1. It can be seen that when we greatly reduce the number of reference points in the fingerprint library, the positioning performance of the present method is still better than that of the conventional adaptive weighting method when the number of reference points in the fingerprint library is 441.
[0095] It is to be understood that the present invention is described by some embodiments, and it is known to those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the scope of protection of the present invention.
Claims
1. A fingerprint library indoor positioning method assisted by intelligent reflective surface, characterized in that: The following steps are involved: Step 1: Create a fingerprint database; Define the line-of-sight link channel between the user and the base station as h d , the channel between the user and the smart reflective surface is h r , the channel between the intelligent reflection surface and the base station is g, and the channel model is a single-path time-invariant channel model, then G = g*diag(h r ) is defined as the cascade channel between the user, the smart reflective surface, and the base station; The base station is equipped with multiple antennas, the user is equipped with a single antenna, the intelligent reflection surface is equipped with multiple reflection units, and multiple fixed coordinate points are pre-selected indoors as reference points of the fingerprint library; the intelligent reflection surface is turned on, the user sends pilots in multiple time slots at each fixed coordinate point, and the base station obtains the channel state information CSI of the total link, which is recorded as CSI-1; the intelligent reflection surface is turned off, the user sends pilots in multiple time slots at each fixed coordinate point, and the channel state information CSI of the link that does not pass through the intelligent reflection surface is obtained, which is recorded as CSI-2; the indirect channel state information CSI value passing through the intelligent reflection surface is obtained by subtracting CSI-1 from CSI-2, which is recorded as CSI-3; the steps of obtaining CSI-1, CSI-2, and CSI-3 are repeated in each time slot, so that the result satisfies the statistical probability distribution; the amplitude information of the above CSI-1, CSI-2 and CSI-3 is obtained, and the link P that does not pass through the intelligent reflection surface is established respectively. d , link P reflected by the smart reflective surface r and the total link P to obtain the channel state information CSI fingerprint library; Step 2: Obtain neural network training samples; Q coordinate positions are randomly selected indoors, and a user sends a pilot signal at each position. The channel state information CSI of the three links mentioned in step 1 is obtained at the base station by switching the intelligent reflective surface, and the three types of fingerprint information in the channel state information CSI fingerprint library obtained in step 1 are subtracted respectively. The obtained differences are squared, and then weighted summed to obtain the coordinate positions of multiple reference points with the largest sum, which are used as the input of the neural network, and the actual coordinate parameters of the point are used as the output; step 2 is repeated Q times to obtain Q groups of neural network training samples; Step 3: Train the neural network; Step 4: In the online stage, the user sends a positioning request to the base station; The base station responds and prepares to receive the pilot sent by the user, and obtains the channel state information CSI of the three links using the method in step 2 and takes the modulus value; Step 5: Match the channel modulus values obtained in step 4 with the fingerprint library to obtain the coordinate positions of the most matching points; The possible coordinate position points are input into the neural network obtained in step 3 for calculation to obtain the estimated position information, and the positioning is completed.
2. The fingerprint library indoor positioning method assisted by intelligent reflective surface according to claim 1 is characterized in that: Step 3 specifically includes the following steps: Perform multiple rounds of neural network training on the xyz three-dimensional space coordinate values, and estimate the actual position information based on the given possible position information; compare the estimated position with the actual position of each point in step 2. If the average positioning error of the test point is less than the threshold D, save the set of neural networks, otherwise repeat this step.
3. The fingerprint library indoor positioning method assisted by intelligent reflective surface according to claim 1 is characterized in that: In step 2, the weight ratio between the channel state information CSI of the link passing through the smart reflective surface, the channel state information CSI of the link not passing through the smart reflective surface, and the channel state information CSI of the total link is not unique. The positioning performance is good when the weight of the link not passing through the smart reflective surface is greater than the sum of the weights of the other two.
4. The fingerprint library indoor positioning method assisted by intelligent reflective surface according to claim 1 is characterized in that: Reducing the number of fingerprint database points in step 1 and the number of training sample groups Q in step 2 can improve the positioning efficiency while maintaining good positioning performance.
5. The fingerprint library indoor positioning method assisted by intelligent reflective surface according to claim 1 is characterized in that: In step 3, the neural network used is a BP neural network or a convolutional neural network CNN.
Citation Information
Patent Citations
Beam domain channel angle estimation method for intelligent reflector-assisted communication
CN112887002A
Channel estimation method for passive intelligent reflection surface based on deep learning
CN113179232A
